Granted Patent B2

US 12,271,695 · App. 17/612,532 · Granted Apr 8, 2025

Inventors: Nitesh Laller (Noida, IN); Sujit Kumar Sinha (Noida, IN)

Assignee: Samsung Electronics Co., Ltd.

G06F40/284G06F40/40G06F40/56G06F40/58G10L15/02G10L15/22G10L15/26G06N20/00

Abstract

Claims (53)

1. A method performed by a user equipment (UE) for reconstructing primary text data, the method comprising:

receiving textual data and a voice input signal corresponding to a user, wherein the textual data includes a textual communication or social network interaction of the user;

generating, based on a detection of words spoken by the user, input text data, wherein the detection of words is based on a feature vector that is indicative of acoustic characteristics of the voice input signal;

generating tokens based on the input text data and the textual data, wherein the tokens include at least one of words, phrases, idioms, jargon, slang, or dialect related to the user;

adding the tokens into a word bag corresponding to the user, wherein the word bag includes at least one of the tokens or labels corresponding to the tokens;

generating (i) a text corpus based on the input text data and (ii) a user profile corresponding to the user based on the word bag, the input text data, and the text corpus, wherein the text corpus includes the input text data and at least one of translations, paraphrases, or contextual labels corresponding to the input text data;

identifying words or phrases from a primary text data or a number of characters in the primary text data, wherein the primary text data includes at least one of media subtitles, virtual assistant output, or machine generated text data;

determining, based on the word bag associated with the user profile and the words or phrases from the primary text data or the number of characters in the primary text data, whether reconstruction of the primary text data is required; and

in case that the reconstruction of the primary text data is required, reconstructing the primary text data to generate a personalized text data based on the user profile, wherein the reconstructing of the primary text data is performed by a natural language processing operation based on a sequence to sequence long short-term memory (LSTM) RNN model that is trained based on the text corpus.

2. The method of claim 1 ,

wherein the labels corresponding to the tokens comprise at least one of language of the tokens, contextual profile corresponding to the tokens, or frequency of use of the tokens by the user.

3. The method of claim 1 , wherein a text reconstruction module is trained for reconstructing the primary text data based on the text corpus.

4. The method of claim 1 , further comprising:

extracting one or more acoustic features from the voice input signal to generate the feature vector;

decoding the feature vector based on a lexicon to detect one or more words spoken by the user;

detecting one or more sentences spoken by the user based on an acoustic model, a linguistic model, or the detected words; and

generating the input text data based on the detected words and sentences.

5. The method of claim 4 , further comprising:

identifying one or more languages of the words and sentences in the input text data;

classifying the words and sentences in the input text data;

labelling the classified words and sentences with corresponding languages; and

generating the tokens based on the classified words and sentences and corresponding labels.

6. The method of claim 5 , further comprising:

translating the sentences in the input text data into one or more other languages to generate one or more translated sentences;

paraphrasing the sentences in the input text data to generate one or more paraphrased sentences;

translating the paraphrased sentences to generate one or more translated paraphrased sentences;

tagging the translated sentences, the paraphrased sentences, and the translated paraphrased sentences with the input text data; and

generating the text corpus, wherein the text corpus includes the input text data and the tagged translated sentences, paraphrased sentences, and translated paraphrased sentences corresponding to the input text data.

7. The method of claim 6 , further comprising:

extracting words and phrases from the primary text data;

performing a lookup for the extracted words and phrases in the tokens in the word bag; and

determining that the primary text data is to be reconstructed when the lookup is below a predefined threshold of lookup.

8. The method of claim 7 , further comprising:

computing a number of characters in the primary text data; and

determining that the primary text data is to be reconstructed when the computed number of characters is more than a predefined threshold of characters.

9. The method of claim 8 , further comprising:

extracting words and phrases from the primary text data;

searching for frequently used words and phrases in the word bag, the frequently used words and phrases corresponding to the extracted words and phrases; and

determining that the primary text data is to be reconstructed based on the frequently used words and phrases in the word bag.

10. The method of claim 1 , further comprising:

receiving the voice input signal and one or more sensory signals; and

identifying a context based on the received voice input signal and one or more sensory signals, the context being indicative of surrounding environment of the user, presence of other people in vicinity of the user, location of the user, and other contextual factors.

11. A user equipment (UE) for reconstructing primary text data, the UE comprising:

a transceiver; and

at least one processor coupled with the transceiver and configured to:

receive textual data and a voice input signal corresponding to a user, wherein the textual data includes a textual communication or social network interaction of the user;

generate, based on a detection of words spoken by the user, input text data, wherein the detection of words is based on a feature vector that is indicative of acoustic characteristics of the voice input signal;

generate tokens based on the input text data and the textual data, wherein the tokens include at least one of words, phrases, idioms, jargon, slang, or dialect related to the user;

add the tokens into a word bag corresponding to the user, wherein the word bag includes at least one of the tokens or labels corresponding to the tokens;

generate (i) a text corpus based on the input text data and (ii) a user profile corresponding to the user based on the word bag, the input text data, and the text corpus, wherein the text corpus includes the input text data and at least one of translations, paraphrases, or contextual labels corresponding to the input text data;

identifying words or phrases from a primary text data or a number of characters in the primary text data, wherein the primary text data includes at least one of media subtitles, virtual assistant output, or machine generated text data;

determine, based on the word bag associated with the user profile and the words or phrases from the primary text data or the number of characters in the primary text data, whether reconstruction of the primary text data is required; and

in case that the reconstruction of the primary text data is required, reconstruct the primary text data to generate a personalized text data based on the user profile, wherein the reconstructing of the primary text data is performed by a natural language processing operation based on a sequence to sequence long short-term memory (LSTM) RNN model which is trained based on the text corpus.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (44)